AI Proxy

This is the start-here guide for AI Proxy on Sealos.

Follow this page when you want the shortest safe path to one first OpenAI-compatible request.

By the end of this page, you will collect the minimum credentials, run one first API call, and know where to confirm that the request reached Sealos.

What AI Proxy is

AI Proxy is the Sealos entry point for making OpenAI-compatible AI requests through one managed endpoint.

Why use AI Proxy

For a first call, AI Proxy gives you one place to get credentials, one request shape to start with, and one place to confirm that the request reached Sealos.

Before You Begin

Use this page if all of the following are true:

• You can sign in to Sealos.
• You can open the AI Proxy product surface in the Sealos console.
• You are ready to copy an endpoint and API key into shell environment variables.
• You will use the fixed cURL example in this page for the first request.

The goal is first success, not full AI Proxy coverage.

Get your endpoint and API key

Collect the two values needed for the first request.

1. Sign in to Sealos and open the AI Proxy surface.
2. Use the current console action to create or reveal an API key.
3. Copy the endpoint shown in AI Proxy.
4. Copy the API key value and keep both values ready for the shell example below.

The next section places those values into shell environment variables.

Make your first API call

Export the three placeholder variables first and run the request second.

export AI_PROXY_BASE_URL="replace-with-the-endpoint-shown-in-ai-proxy"
export AI_PROXY_API_KEY="replace-with-your-api-key-value"
export AI_PROXY_MODEL="replace-with-a-model-available-in-your-workspace"
 
curl "$AI_PROXY_BASE_URL/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AI_PROXY_API_KEY" \
  -d '{
    "model": "'"$AI_PROXY_MODEL"'",
    "messages": [
      {"role": "system", "content": "You are a concise assistant."},
      {"role": "user", "content": "Reply with one short sentence about Sealos."}
    ],
    "temperature": 0
  }'

Replace the three placeholders with the values you copied from Sealos.

A successful response should return generated text in the response body.

The next section explains only the three fields used in this request.

Understand the request fields

These are the only fields you need to understand for the first call:

• model selects the AI model the request should use. This page keeps the starter model neutral until later verification.
• messages carries the conversation input for the request. The first request keeps it minimal so the response is easy to recognize.
• temperature controls response randomness. This first request keeps it low and simple for a predictable first success.

Check logs and usage

Check these in order after the first call:

• Confirm the request returned successfully and the response contains recognizable generated text.
• Find the same request record in AI Proxy logs. This is the primary Sealos-side signal.
• Use usage or billing visibility as supporting confirmation only.

The next section covers only the first set of failure checks.

If the first call fails

Check these first:

• wrong endpoint root or wrong path Look here first: keep the AI Proxy endpoint unchanged and confirm the request still targets /v1/chat/completions.
• wrong API key Look here first: re-copy the current key and confirm the request still sends Authorization: Bearer $AI_PROXY_API_KEY.
• unsupported or unavailable model Look here first: replace the placeholder with a model that is available in your workspace.
• no log record Look here first: if the request does not appear in AI Proxy logs, it may not have reached AI Proxy at all.
• missing balance or usage visibility Look here first: if the request reaches AI Proxy but does not complete, confirm the workspace still has usable balance or visible usage data.

What this page does not cover

This page does not cover a model catalog, deep billing or deep logs guidance, or advanced proxy features such as caching, routing, quotas, or failover.

It is the shortest path to a first successful call.